mk-sign-language-real-time-recognition

Image annotation -> I did this localy, it is not tested in Colab

1. pip install -r requirements.txt
2. Use Image Collection.ipynb to take the images
3. Move them to workspace/images/colectedimages
4. Go to labelImg folder
5. python labelImg.py
6. Use Open Dir and select workspace/images/colectedimages
7. Change Save Dir to workspace/images/colectedimages
8. Annotate the images
9. Go to workspace/images and run prepare-train-test.ipynb

Label IMG Shortcuts

Ctrl + u - Load all of the images from a directory

Ctrl + r - Change the default annotation target dir

Ctrl + s - Save

w - Create a rect box

d - Next image

a - Previous image

del - Delete the selected rect box

Ctrl++ - Zoom in

Ctrl-- - Zoom out

Ctrl + d - Copy the current label and rect box

Space - Flag the current image as verified

↑→↓←Keyboard arrows to move selected rect box

Image augmentation

1. Image augmentation done with imgaug
2. Link to augmentation notebook

Model train -> Do this in Colab

1. Run pipeline.ipynb in colab env
2. There are a few paths that need to be adjusted to the directory in which you will clone the repo
3. Test it with predictions.ipynb

Predictions -> Do this in Colab

1. Run predictions.ipynb in Colab env after you completed pipeline.ipynb

To-do

1. work out the real time solution in predictions.ipynb -> Done
2. Train on more images -> Done
3. Write documentation overleaf -> Done

Real-time preview

realtime_obj_detection_git.mov